I think altmans code red included specifically training for your pelican test to get positive coverage from the community.

looks also a lot sexier in dark

PS: also we would you consider production code gen for testing not a good test case? (as long as you review the code)

for further testing, you might want to consider esoteric programming languages, or even making up your own syntax, to better indicate if the model employes really generalist reasoning and not just in data distribution optimization.

Sorry no.

Honestly, that’s one of the worst analogies I heard. It adds nothing of value to the old analogy of the managerial position, just placed it in a unnecessary domain specific role which makes it more confusing.

Can you post specific instances of breaches ?

For many tasks, there are model-capability thresholds that can already be saturated (e.g. SQL query generation, various fiction texts), so the future of LMs are in many ways personalised, private, on-device models giving rise of independence of the big LLM providers and their data harvesting.

need uniforms so they feel strong

so what tools were used?

Also the discussion around it, as i think universites also fall under that small but important organizations banner

news.ycombinator.com/item?id=4536...

I think this would also interest you
Regarding power dynamic shifts bu way of decrease of small organizations
mathstodon.xyz/@tao/1152599...

And start calling them text predictors

People are so tired of big tech companies messing with their apps / UIs to up-worsen and shovel new (paid) upgrades to their face.

Looks interesting indeed, looking forward to Simon’s notes on it.

Microsoft being Microsoft again

Cool, how much is the whole setup?

Feels like models (local or API) are being held back by their UI currently…need more than a chat window, need to be able to write and organize projects (so perhaps note taking app, obsidian like) but also need to run/validate its code so IDE like. Perhaps a UI that can flick between those views.